International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
evaluation of semi-automatic road extraction from aerial images. 
This buffer method was modified for our purpose to assess the 
quality of the delineated stand borders. 
The principle of the buffer-method is described in Fig. 2. 
Fig. 2. Matching principle of the delineated stand borders and 
the reference geometry: buffer method after Heipke et 
al. (1998). 
In the first step, a buffer of constant predefined width (buffer 
width) is constructed around the reference forest inventory 
border (dotted box). The parts of the delineated geometry 
(dotted line) within the buffer are considered as matched (see 
Figure 2a). The matched extracted data are denoted as true 
positive with length TP. The unmatched delineated data are 
denoted as false positive with the length FP. 
In the second step, matching is performed the other way round. 
The buffer is now placed around the delineated border, and the 
parts of the reference data lying within the buffer are considered 
to be matched and called true negative with length TN (see 
Figure 2b). The unmatched reference data are denoted as false 
negative with the length FN. 
Heipke et al. suggest to calculate this quality measure with the 
intention to compare the results of different road extraction 
methods, not as a perfect solution for quality assessment. As 
quality measure, the authors defined: 
Completeness = length of matched reference 
length of reference 
= TN / (TN + FN), optimum value is 1 
Correctness = length of matched delineation 
length of delineation 
= TP / (TP + FP), optimum value is 1 
Quality = length of matched delineation 
length of delineation + length of unmatched ref. 
= TP / (TP + FP + FN), optimum value is 1 
For the above mentioned purpose, correctness is the most 
important quality measure: it represents the percentage of 
correctly delineated forest stand border, explicitly the percentage 
of the delineated stand border within the buffer placed around 
the reference. The quality is a more general measure of the final 
result giving a combination of completeness and correctness in 
one single expression. 
3. RESULTS 
All three sensor fusion methods (IHS, PCA, Brovey) showed a 
synergy effect by combing high spectral and spatial satellite 
data. The integration of high resolution data increased the 
visible interpretability of multispectral data for updating forest 
maps. This is caused by the improved recognition of linear 
features like logging roads, stand borders and also textural 
patterns. 
For visual interpretation, the IHS transformation showed the 
best colour differentiation (Figure 3). The visual impression of 
the PCA transformation is similar to the IHS transformation. 
Coniferous and deciduous trees can be detected very clearly. 
Even clear-cuts appear in violet-red colours. 
The result of the Brovey transformation showed comparatively 
poor differentiation in brightness and tone. Although the 
colouring is similar to a NIR image, the interpretability of 
different forest types and age classes is more difficult. This is 
also due to the high spectral resolution of the TM with three IR 
bands. 
Fig. 3. Fused images of KVR and AIF with 2m resolution. 
Due to the good performance of the IHS transformation, the 
applicability of sensor fusion techniques for forest inventory 
mapping has been investigated with an IHS transformed product 
of Landsat TM and IRS-1C pan [IHSTM], This product was 
compared to an IHS transformed SPOT XS and PAN [IHS SP], 
IRS-1C pan [Pan] alone, a B/W orthophoto [Ortho] and a 
simulated QuickBird image [Qsim] with lm resolution (see 
Table 4). 
3.1. Visibility percentage 
The first analysis step was to measure the visibility percentage 
of forest stand borders in comparison to the official forest